We bring you another recap of all the AMD Radeon RX 7000 , their performance comparisons and the news of the RDNA 3 architecture in the RX 7900 XT, 7800 XT, 7700 XT and 7600 XT.
AMD improves year after year, and it couldn’t be any other way than with TSMC as an ally. Thanks to its 5 nm, we have been able to see an upgrade compared to the RX 6000, but the start of this generation of graphics cards has been somewhat abrupt. They have passed the occasional GPU through our test bench, so we took the opportunity to collect all the information in a single article.
AMD Radeon RX 7000.
Index of contents
- AMD Radeon RX 7000: all the news
- RDNA 3, the heart of the RX 7000
- 5 nm process: more performance / watt
- The AMD Radeon RX 7000, prepared for a frequency of +3 GHz
- More and better Compute Units
- IA units are AI accelerators
- New AI instructions to speed up Ray Tracing
- New Infinity Cache
- 2nd generation Ray Accelerators
- AMD Radiance Display y DisplayPort 2.1
- AV1 codec for using AMD Radeon RX 7000 in streaming
- AMD Smart Access Video
- Benchmark comparison (pending update)
- Test bench and data source
- Fire Strike y Time Spy
- VRMark Orange Room y Port Royal
- Pure Ray Tracing
- Test gaming (pending update)
- 1080p sin y con Ray Tracing
- 1440p sin y con Ray Tracing
- 2160p sin y con Ray Tracing
- averages
- AMD Radeon RX 7000 prices
- Last conclusions
AMD Radeon RX 7000: all the news
We are going to start by presenting these magnificent graphics cards that have respected the price and the pocket of consumers. So let’s see what AMD has brought us with this 7000 series.
RDNA 3, the heart of the RX 7000
The central pillar of this generation of AMD Radeon is RDNA 3 , to which we already did its corresponding review by our colleague Isaac. Broadly speaking, AMD has opted for a chiplet design, as is the case with its Zen architecture in Ryzen, which is historical in graphics cards.
This means that we will have 2 key elements: the GCD (Graphics Compute Die) and the MCD (Memory Cache Dies). Inside the GCD we find the CU, the ALU, SIMD and the MCD; likewise, AMD has included IA drives and 2nd generation Ray Accelerators. Inside the MCDs are the L3 cache blocks and the GDDR6 memory interface. For example, inside a Navi 31 XT we have 1 GCD with 6 MCD inside.
The GCD is manufactured in a 5 nm process , while the MCD is manufactured at 6 nm , each measuring 300 mm² and 37 mm² respectively. In the end we have Navi 3X GPUs that come in a 5nm process, and AMD has highlighted the MCM architecture in their presentations.
All of this is obviously reminiscent of AMD Ryzen CCDs and CCXs on Zen 2 and Zen 3 architectures, along with an IOD that came in a different manufacturing process. So, it seems that they have taken that design or approach to dump it into a GPU, and I can tell you in advance that the move has not gone bad at all.
5 nm process: more performance / watt
As AMD and NVIDIA have advanced lithography, slides that have rated the “performance / watt” ratio are present in each release they announce. The idea is simple: increase the performance of the node by being more efficient than the previous one, but in practice current GPUs consume more power.
To give you an idea of what the AMD Radeon RX 7900 and 7900 XTX consume, we offer you this comparison made by the Professional Review team.
As you can see, we are going to 509 W in the whole team , alerting José Antonio Castillo that if we stress the CPU, we would get into a consumption of 710 W. After several tests, he concluded that the AORUS RX 7900 XTX had an average consumption of 42 W at rest and 409 W on load.
Therefore, in the case of the AMD Radeon RX 7900 XT and the RX 7900 XTX, we are talking about the following consumption increases compared to the RX 6900 XT and RX 6950 XT:
- Increased consumption at rest (IDLE):
-
- RX 7900 XT: 45.21%.
- RX 7900 XTX: 43.48%.
- Increase in load consumption:
-
- RX 7900 XT: 3.68%
- RX 7900 XTX: 3.67%.
-
-
At rest is where consumption increases the most, but we must say that AMD has done fantastically under load.
The AMD Radeon RX 7000, prepared for a frequency of +3 GHz
That was announced by AMD in the press slides, but has it really come true? In the AMD Radeon RX 7900 XT review , we saw José bump the clock up to 2930 MHz , stating that there are no FPS improvements beyond 2.9 GHz: barely 2 FPS in 2K and 4K.
We also tested the AORUS RX 7900 XTX, where we saw that GIGABYTE increased 200 MHz without reaching the actual 3 GHz, improving few FPS in practice. Anticipate that the custom GPUs are going to arrive very adjusted from the factory so that practically nothing can be done to them.
With all this, AMD claims that RDNA 3 has been designed to reach 3 GHz , but the Boost frequencies are below as we have told you. Surely, they can be achieved with liquid cooling or a model comes to the market that is capable; by air, it will be very complicated and inefficient.
In fact, the red team claims that it can run at the same frequency as RDNA 2 using half the power.
More and better Compute Units
Let’s start with the data that AMD gives:
Enhanced CUs provide approximately a 17.4% clock-for-clock architectural improvement.
We’re going from 80 CU to 96 CU , which is a 20% increase in these shaders, but let’s not stop at quantity. You already know that within the CU we find the Stream Processors , something similar to the NVIDIA CUDA Cores. The first reports indicated that we had 128 Stream Processors per CU, which gave double the performance of FP32 when compared to RDNA 2.
Some uncertainty was planted in the presentation because we saw the Stream Processors separated from the CUs, resulting in giving different total figures. Earlier we said that each CU had 128 Stream Processors, and the RX 7900 XTX has 96: 12,288 Stream Processors.
In the end, in the technical sheets we see 6144 shaders, so we recommend staying with this last figure because it seems to be the consensus. AMD halved the total shaders that the RX 7000 has, but it’s a superficial change, so it was more a question of distribution of numbers, than power.
The new CUs share resources between rendering, artificial intelligence and Ray Tracing to make effective use of each transistor, as well as provide more performance to these graphics cards.
Finally, each CU contains 4 Texture units , 1 Ray Accelerator and 2 IA Units , with the 96 CUs of the RX 7900 XTX divided into 6 shader engines. The best we can do is buy the RX 7000 vs RX 6000 shaders to appreciate it better.
| AMD Radeon RX 7000 | Shaders | TMUs | ROPs |
| RX 6600 / XT | 1792 / 2048 | 112 / 128 | 64 / 64 |
| RX 7600 / XT | 2048 / N/A | 128 / N/A | 64 / N/A |
| RX 6700 / XT | 2304 / 2560 | 144 / 160 | 64 / 64 |
| RX 7700 / 7700 XT | N/A | N/A | N/A |
| RX 6800 / XT | 3840 / 4608 | 240 / 288 | 96 / 128 |
| RX 7800 / XT | N/A | N/A | N/A |
| RX 6900 XT / 6950 XT (solo mejor VRAM) | 5120 / 5120 | 320 / 320 | 128 / 128 |
| RX 7900 XT / XTX | 5376 / 6144 | 336 /384 | 192/192 |
There are still a few model releases for the AMD Radeon RX 7000 for which we have very little information. Looking at how AMD scales shaders, TMUs, and ROPs, expect the RX 7600 XT to have 2,500+ shaders, the RX 7700 XT to come with 3,000+ shaders, and the RX 7800 XT to come close to 5,000 shaders.
In fact, it appears that the non-XT models have the same shaders as the previous generation XT models . This table will be updated as we see announcements and releases, so we ask for your patience.
IA units are AI accelerators
This is a super important change in RDNA 3, but AMD has not specified how they are incorporated into the architecture, simply that we have 2 per CU, that there are “new instructions” and that their IA performance has been improved, communicating with the CU via 2 SIMD.
Everything indicates that the AI accelerators share the resources and registers with the ALUs , and that is that AMD aims for a super high figure, almost triple the performance! Many analysts have spoken about it and assure that this data is possible thanks to the increase in the frequency of the GPU, as well as the increase in CU.
We’ll just let you know what each other says, as we have no way of proving the source of the reported performance increases.
New AI instructions to speed up Ray Tracing
In the end, you have to surrender to the evidence provided by artificial intelligence for calculations as heavy as those resulting from Ray Tracing. AMD would not only have thought to include the AI accelerators and forget about the rest, but has incorporated new IA instructions in RDNA 3 to finish off the job.
RDNA 3 has brought with it the Wave Matrix Multiply Accumulate (WWMA) feature , a feature that promises to deliver hardware-accelerated matrix multiplications in video games, said Aaryaman Vasishta . Mike Mantor, the Head of AMD GPU Architecture stated the following:
The WMMA instruction optimizes data movement scheduling and maximum math operations with minimal VGPR access by providing source data reuse and intermediate destination data forwarding operations without interruption. The regular patterns experienced in matrix operations allow the WMMA instructions to reduce the required power while still providing optimal operations that allow sustained operations at or near maximum rates.
In this way, this was the plan that AMD had to improve its performance in Ray Tracing, although it remains NVIDIA territory today.
New Infinity Cache
Be careful because AMD has lowered the amount of L3 cache in RDNA 3 , but the brand bases it as a real improved use: it is possible that the GPU can store data and access it several times. In other words, consumption would have been reduced (it takes less L3) and agility is improved by eliminating traffic to VRAM .
We must say that the communication between GCD and MCD is vital for bandwidth, and in this sense AMD ensures that a peak bandwidth has been improved almost 3 times. Thus, here we will see 5.3 TB/s of bandwidth .
2nd generation Ray Accelerators
The third in contention are the RA, some units that are inside the CU, but that have been improved in the AMD Radeon RX 7000. Apparently they have improved performance by up to 50% per CU through the instructions that we have already mentioned.
Not everything depends on adding some new instructions, but the Ray Accelerators have incorporated improvements in classification and analysis in the filtering of the ray boxes or “ray boxes”. The key is the BVH method for calculating Ray Tracing because it is used in games compatible with this technology, which are released today.
AMD Radiance Display y DisplayPort 2.1
We did not expect this novelty in AMD Radeon RX 7000, but the truth is that the video and multimedia outputs have worked through an engine called AMD Radiance Display . This engine offers 12-bit per color channel , reaching 68 billion colors. It is characterized by its compatibility with DisplayPort 2.1 , thus guaranteeing the best possible support for 8K and a refresh rate that comes from the future.
Yes, 8K at 165Hz and 4K at 480FPS is the maximum that DisplayPort 2.1 offers, which seems like a real outrage to us. Right off the bat, we went looking for DisplayPort 2.1 monitors on the market to see which ones could really squeeze the RX 7000. ASUS introduced the ROG Swift PG32UQXR in January as the first ROG monitor with DisplayPort 2.1, which is an evolution of the current PG32UQX, whose price is not a bargain. Samsung also announced an Odyssey Neo G9 … but neither are on the market.
The point is that we will be able to reach 900 Hz in QHD , 480 Hz in 4K and 165 Hz in 8K , being a brutal upgrade from the RX 6000.
Who knows if in the future this is feasible with a totally futuristic FSR version?
AV1 codec for using AMD Radeon RX 7000 in streaming
Although in Professional Review we already talked about the AV1 codec , NVIDIA and AMD have focused on incorporating this new acceleration for streamers and content creators. They claim that it is up to 7 times faster decoding 8K in FFmpeg. Specifically, AMD RDNA 3 Media Engine has 2 engines (Dual Media Engine) to simultaneously process 2 independent encoding and decoding streams.
AMD Smart Access Video
What is AMD SmartAccess Video ? Well, it seems to be a technology whereby the CPU and GPU are synchronized to improve hardware encoding and/or decoding performance. It is intended for when many simultaneous streams appear on platforms with Ryzen 7000 and AMD Radeon RX 7000, but without being the Smart Access Memory that we saw before.
According to AMD, the idea is that it will be used for powerful professional applications, as well as for those used for streaming.
Benchmark comparison (pending update)
As of May 28, we only have 3 AMD RX 7000 graphics cards that we have been able to review.
Test bench and data source
As usual, the data collected is extracted from the individual reviews of each GPU, having the same platform in all.
| TESTING BENCH | |
| Processor: | Intel Core i9-12900K |
| Base Plate : | Asus ROG Strix Z790 Hero |
| RAM memory: | 32 GB Kingston Fury Beast DDR5 5600MHz |
| heatsink | Corsair H150i Elite LCD |
| HDD | Samsung 860 QVO |
| Graphic card | AMD Radeon RX 7600
AMD Radeon RX 7900 XT AMD Radeon RX 7900 XTX |
| PSU | Corsair RM1000 |
| Monitor | Viewsonic VX3211 4K mhd |
As the RX 7000 passes through our test bench, we will update the results of the comparisons. At the moment, only the RX 7900 XT, XTX and the RX 7600 have passed, leaving clear differences between them.
Fire Strike y Time Spy
Starting with the Fire Strike and Time Spy tests, we draw these conclusions:
- In the Extreme and Standard tests we see a fairly normal difference between the RX 7900, but in Ultra it is reduced to almost half.
- Time Spy is better for lower GPUs, but the RX 7600 is still light years ahead of the RX 7900, it feels like it’s from another generation!
VRMark Orange Room y Port Royal
Curious at least the objective result in VRMark Orange Room, seeing an RX 7600 even ahead of the RX 7900 XT . Normalcy is back with Port Royal, and it looks like the XTX is going to be far superior to the RX 7900 XT, sure? Go placing bets.
Pure Ray Tracing
In my opinion, this test leaves the AMD Radeon RX 7000 as good graphics cards for ray tracing, but enough to pick them over NVIDIA? Comment below what you think.
Definitely, the RX 7600 does not seem to be very comfortable in this scenario.
Test gaming (pending update)
Let’s go with what’s cool, and that is that we have tested the games with these graphic settings:
- Shadow of the Tomb Raider, Alto, TAA, DX12
- Far Cry 5, Alto, TAA, DX12
- Eternal DOOM, Ultra, TAA, Volcano
- Final Fantasy XV, standard, TAA, DX11
- Deus EX Mankind Divided, Alto, DX11
- Metro Exodus, Alto, DX12
- Control, High, no RTX, 1080p rendering, DX12
- Gears 5, Alto, TAA, DX12
- Red Dead Redemption 2, Alto, DX12
- Horizon Zero Dawn, Calidad, DX12
- Assassins Creed Valhalla, Alto, DX12
With Ray Tracing, the following:
- Shadow of the Tomb Raider, RT Alto, XeSS Equilibrado, DX12.
- Metro Exodus, RT Alto, DX12.
- Control, RT High, rendered at 50%, DX12.
- Call Of Duty Modern Warfare, RT Alto, DX12.
- Battlefield V, RT Alto, DX12.
- Cyberpunk 2077, RT Ultra, FSR Performance, DX12.
1080p sin y con Ray Tracing
They all top 100 FPS with no issues in this test, though the RX 7600 barely does it in Metro Exodus. I would like to point out that the RX 7900 XT achieves better results in Tomb Raider, Far Cry 5, Final Fantasy XV and Gears 5: in half the games it is superior to the RX 7900 XTX.
The little difference between the 3 in Far Cry 5 is striking , with the RX 7600 being a GPU 3 ranks smaller than the RX 7900. We are looking forward to seeing how the RX 7700 and 7800 XT perform.
In Ray Tracing comes the catastrophe for the RX 7600 , a GPU that already tells us that it is coming for 1080p at most in this scenario. The results obtained by the RX 7900 XT and XTX are, in my opinion, disappointing in Cyberpunk 2077, barely exceeding 100 FPS.
Battlefield V is a poorly optimized game, so we have to be fair to the GPUs. Very good performance in CoD, Control, Metro Exodus and Tomb Raider.
1440p sin y con Ray Tracing
The 1440p review starts with a rather feisty RX 7600 without Ray Tracing , which we recommend for QHD and 60 FPS. At the top of the range of AMD Radeon RX 7000, we find some RX 7900 XTX and XT with fabulous performance, as well as a new victory in Gears 5 of the little sister.
The good news here is that we will be able to play Ray Traced with full ultra and 1440p at 60 FPS on the RX 7900 XT and XTX . Conversely, don’t expect that possibility on the RX 7700 and 7800 XT.
The RX 7600 manages to exceed 60 FPS in Control and CoD, with the most powerful RDNA 3 GPUs being interesting for 144 Hz monitors.
2160p sin y con Ray Tracing
We end up with 4K , the target resolution of the AMD RX 7900 XT and 7900 XTX . Just below we will give you the averages, but let me tell you that these RDNA 3 GPUs are interesting for a 60 Hz and even 120 Hz 4K monitor.
Unsurprisingly, the RX 7600 falls short of 60 FPS in most games, with Control and Doom Eternal being two very well-optimized exceptions on AMD.
We are going to finish the 4K and RTX ON gaming tests with very different results: we can play more than 120 FPS in CoD, Control or Tomb Raider with the best, but it will be impossible to reach 60 FPS in powerful games.
averages
After briefly analyzing the results, we go with the averages to give you a general idea of the gaming performance and the FPS that the AMD Radeon RX 7000 can get .
| 1080p / RTX | 1440p / RTX | 2160p / RTX | |
| RX 7600 | 148.1 / 78.67 FPS | 103.6 / 56.83 FPS | 58.8 / 29.17 FPS |
| RX 7900 XT | 240.6 / 164.83 FPS | 210.8 / 133.67 FPS | 134.1 / 80.5 FPS |
| RX 7900 XTX | 249.5 / 182.83 FPS | 227.4 / 151.83 FPS | 160.1 / 95.3 FPS |
As a curiosity, I wanted to calculate the percentage loss of FPS RTX OFF/ON:
| 1080p / RTX | 1440p / RTX | 2160p / RTX | |
| RX 7600 | 53.12% | 54.86% | 49.61% |
| RX 7900 XT | 68.51% | 63.41% | 60.03% |
| RX 7900 XTX | 73.28% | 66.77% | 59.53% |
AMD Radeon RX 7000 prices
We are going to take the opportunity to put what we know now and, thus, monitor prices from the date of departure of the graphics cards.
| RX 7600 | RX 7900 XT | RX 7900 XTX | |
| starting price | From €299 | From €1090 | From €1,190 |
| Price as of 5/30/2023 | From €299.99 | From €949 | From €1098.99 |
Last conclusions
After the extensive analysis we have conducted, we have drawn several direct conclusions.
First of all, we are saddened to see how late AMD is in launching mid-range and high-end models, since they are the best-selling models of the brand in the previous RX 6000. In the middle of 2023, they have only released 2 very high-end GPUs that They are above €1000 , but we see that they are going down in price.
Secondly, we see that performance and performance-to-watt ratio have improved a lot , which is a huge leap from RDNA 2. Still, it seems that NVIDIA’s alternatives are still better.
Going to the third point, prices have risen almost €100 starting from one generation to another and we are seeing GPUs at a price that people cannot afford. The consequence of this is that we continue to see RTX 3000 and RX 6000 on the market: this must change.
The fourth conclusion has to do with consumption, and it is that it has been greatly reduced thanks to 5 nm. Of course, all in exchange for a very high price, and we will need to have modern processors so as not to create a bottleneck.
Regarding performance, the Ray Tracing of these graphics cards is still somewhat disappointing, even though they have improved FidelityFX Super Resolution . And speaking of FSR , back in June we are still waiting for FSR 3, which was released alongside RDNA 3.
To be fair, we must positively highlight the performance of the RX 7600, but they do not give it to us: from €299 a GPU that previously cost you around €230 to €250 . I’m not talking about performance, but about its position in the AMD Radeon RX 7000 range compared to its older sisters.
Last year we did a comparison of RX 6600 vs RX 6600 XT vs RTX 3060 and 3060 Ti , where we saw that the RX 6600 could not compete with the RTX 3060; In fact, if we went to 1440p or 2160p, the RX 6600 XT lost against the RTX 3060. Not to mention Ray Tracing.
We understand that since the RTX 2000 came out in 2018, we cannot expect that in 2023 we will see the same prices. The REAL mid-range today starts at €400, not €300 as in the past, unfortunately.
We recommend the best graphics cards on the market .
Lastly, and as a personal criticism of AMD and NVIDIA, if we release graphics cards every 2 years, we should enjoy them as users for 2 years . Before, they announced the main squad and then released the Super or Ti versions; today, we have to wait an average of 6 months from the first presentation to play something in the midrange.
We hope that the analysis has been helpful to you. If you have any questions, you can comment below and we will attend to you immediately. Which RX 7000 do you prefer?